Model Dermatology is freely available for academic purpose.
Currently, 30~40 users/day visit the DEMO, and all users are allowed to test the algorithm with a warning message.
Only the user with the email address of academic domains (.ac, .edu, .org, and other university domains of https://github.com/Hipo/university-domains-list) and the email address identified by pubmed are authorized. The submitted image is transferred and saved for research purpose, but, we do not save the image of the authorized users. In case of the server down of the authorization, it also does not save.
Once after you login in this site, you can use the web DEMO for several days if your IP address will not be changed.
The algorithm was developed and maintained by Han Seung Seog, a dermatologist at I Dermatology Clinic (firstname.lastname@example.org).
Wrong predictions for common allergic disorders on the face
Lower accuracy for non-Asians
Lower accuracy for inadequate quality photographs
Inaccurate prediction for the normal lesion of scalp
In the Build2020, there was some improvement for these issues. The Top-1 accuracy for the Edinburgh dataset (1300 images; mainly white population) improved from 53.6% to 60.1%.